drift

Analyzes codebases across 8+ languages (TypeScript, Python, C#, Java, PHP, Go, Rust, C++) using a Rust-based core engine that performs AST parsing and structural analysis to identify recurring patterns, naming conventions, architectural styles, and anti-patterns. Returns pattern matches with statistical confidence scores derived from frequency analysis across the codebase, enabling AI assistants to understand project-specific conventions with quantified certainty rather than guessing.

Maintains a file-system-backed decision store (stored in .drift/ directory) that records architectural decisions, design choices, and conventions made across coding sessions. The memory system allows developers and AI assistants to query previous decisions via MCP, enabling context to persist across IDE restarts and multiple AI interactions without requiring manual re-explanation of project decisions.

Exposes Drift's pattern detection and decision memory capabilities as an MCP (Model Context Protocol) server that integrates directly into IDEs like VS Code and Cursor. The MCP server implements standard tool-calling interfaces allowing AI assistants running in the IDE to query codebase patterns and decisions without leaving the editor, with results automatically injected into the AI's context window for code generation.

Provides a command-line interface (drift init, drift scan, drift import, drift memory) that performs batch analysis of codebases without requiring IDE integration or cloud connectivity. The CLI invokes the Rust core engine to parse and analyze code, stores results in the local .drift/ directory, and outputs human-readable reports or JSON data for integration into CI/CD pipelines and automation workflows.

Analyzes code structure using Abstract Syntax Trees (ASTs) for each supported language, enabling detection of language-specific conventions like naming patterns (camelCase vs snake_case), architectural styles (MVC, layered, modular), and language idioms. The Rust core engine maintains separate parsers for each language, allowing it to understand semantic structure beyond simple text matching and detect violations of language-specific best practices.

Provides a drift import command that allows developers to import existing architectural decisions, patterns, and conventions from legacy documentation, previous analysis tools, or manual records into Drift's persistent memory system. This enables teams to bootstrap Drift with existing knowledge rather than starting from scratch, and facilitates migration from other codebase intelligence tools.

Supports project-level configuration (via .driftrc or similar config files) that allows developers to customize which files/directories are analyzed, which patterns to detect, which languages to prioritize, and how to weight different pattern types. The configuration system integrates with .gitignore for automatic exclusion of ignored files, reducing noise and focusing analysis on relevant code.

Implements a three-tier architecture where performance-critical operations (AST parsing, pattern matching, statistical analysis) run in Rust for speed and memory efficiency, while user-facing interfaces (CLI, MCP server, configuration handling) are implemented in TypeScript for rapid development and Node.js ecosystem access. Native bindings bridge the Rust core and TypeScript interfaces, enabling both performance and accessibility without sacrificing either.

Assigns statistical confidence scores to detected patterns based on frequency analysis across the codebase, indicating how consistently a pattern is followed. Patterns detected in 90% of relevant code receive higher confidence than patterns found in 30% of code, allowing AI assistants to distinguish between established conventions (high confidence) and emerging or inconsistent patterns (low confidence) when deciding whether to apply them to new code.